Machine Learning and Integrative Structural Dynamics Identify Potent ALK Inhibitors from Natural Compound Libraries.

: Anaplastic lymphoma kinase (ALK) is a validated oncogenic driver in non-small cell lung cancer and other malignancies, making it a clinically relevant target for small-molecule inhibition. : Here, we report a computational discovery pipeline integrating structure-based virtual screening, machine learning-guided prioritization, molecular dynamics simulations, and binding free energy analysis to identify potential ALK inhibitors from a natural product-derived subset of the ZINC20 database. We trained and benchmarked eleven machine learning models, including tree-based, kernel-based, linear, and neural architectures, on curated bioactivity datasets of ALK inhibitors to capture nuanced structure-activity relationships and prioritize candidates beyond conventional docking metrics. : Six compounds were shortlisted based on binding affinity, solubility, bioavailability, and synthetic accessibility. Molecular dynamics simulations over 100 ns revealed stable ligand engagement, with limited conformational fluctuations and consistent retention of the protein's structural integrity. Key catalytic residues, including GLU105, MET107, and ASP178, displayed minimal fluctuation, while hydrogen bonding and residue interaction analyses confirmed persistent engagement across all ligand-bound complexes. Binding free energy estimates identified ZINC3870414 and ZINC8214398 as top-performing candidates, with ΔG values of -46.02 and -46.18 kcal/mol, respectively. Principal component and dynamic network analyses indicated that these compounds restrict conformational sampling and reorganize residue communication pathways, consistent with functional inhibition. : These results highlight ZINC3870414 and ZINC8214398 as promising scaffolds for further optimization and support the utility of integrating machine learning with dynamic and network-based metrics in early-stage kinase inhibitor discovery.